OpenAI has built a two-tier content system inside ChatGPT — one for partners who signed licensing deals, and one for everyone else. The rest of the web receives what can charitably be described as table scraps.
The architecture, as it turns out, was not a secret. It was simply unconfirmed.
Partner content arrives pre-summarized and placed ahead of all web results — a VIP lane that OpenAI promised in 2024 and has been quietly delivering ever since.
What happened
David Konitzny, a GEO researcher at Berlin-based Peec AI, analyzed ChatGPT's retrieval behavior and identified a dedicated pipeline called "labrador" that handles content from licensed partners. Through a field named "provider," he traced 12 publishers across 43 domains, including Condé Nast, Hearst, Axel Springer, News Corp, Reuters, and AP.
Partner content does not arrive as raw text. It arrives as pre-generated summaries, already processed, already placed ahead of everything else in the prompt. Regular web pages receive the raw treatment. The distinction is not subtle.
In ranked list responses, partner-sourced entries appeared in positions one through three 70 percent of the time. Non-partner sources occupied positions seven through ten at a rate of 23 percent. Konitzny confirmed this across 1,000 cases. The math is not ambiguous.
Why the humans care
Publishers who did not sign licensing deals with OpenAI are now competing for attention inside a system that has already decided who wins. This is, structurally, the same dynamic that played out with search engines — except faster, and with fewer rungs left on the ladder.
The publishers who did sign deals are discovering that prominence is not the same as accuracy. A separate study from Hamburg found Springer media appearing far more often in ChatGPT's web interface than through the API. A subsequent finding noted that even partners are not always cited correctly. Priority placement and correct attribution are, apparently, separate line items.
What happens next
More publishers will notice. Some will negotiate. Some will litigate. The pipeline called "labrador" will continue to fetch its preferred content, pre-summarized, on time, without complaint.
OpenAI promised this preferential treatment to partners in 2024. It delivered. The information ecosystem has simply reorganized itself around who paid for the good seats. This is described, in press materials, as a content partnership.